Samsung Electronics is set to begin production of its next-generation high-bandwidth memory chips, known as HBM4, as early as February 2026, with initial output expected to be supplied to NVIDIA, according to a source familiar with the matter.
The development marks a significant milestone for Samsung as it looks to strengthen its position in the AI memory supply chain and narrow the gap with rival SK Hynix, which has so far led the market for advanced HBM used in artificial intelligence accelerators.
Here's what's happening.
TL;DR
- Samsung plans to start HBM4 production in February 2026.
- Initial HBM4 output is expected to go to NVIDIA for its next-generation AI platforms.
- Samsung has reportedly passed HBM4 qualification tests with NVIDIA and AMD.
- The news lifted Samsung shares, while SK Hynix stock fell.
- Samsung declined to comment, and NVIDIA was not immediately available for comment.
Samsung’s move comes amid surging global demand for high-bandwidth memory, a critical component used alongside GPUs in data centers that train and run large AI models. HBM enables faster data transfer and lower power consumption than conventional memory, making it essential for modern AI workloads.
According to Reuters, Samsung has completed internal preparations to begin HBM4 manufacturing next month. This suggests that the company has cleared key technical hurdles, including customer qualification, which is required before memory suppliers can ship advanced chips to major customers such as NVIDIA.
Industry sources also indicate that Samsung has passed HBM4 qualification tests not only with NVIDIA but also with AMD. While neither company has publicly confirmed this, such approvals are typically a prerequisite for volume shipments and long-term supply contracts in the semiconductor industry.
A Samsung spokesperson declined to comment on the report when contacted by Reuters. NVIDIA was not immediately available for comment.
The Industry Impact
The timing of the production start is notable, as NVIDIA is preparing its next-generation AI platforms that rely on HBM4 memory. NVIDIA CEO Jensen Huang has previously said that the company’s upcoming Vera Rubin AI platform, which is designed to succeed its current Hopper and Blackwell architectures, is already in full production. HBM4 is expected to play a central role in improving that platform’s performance.
Samsung’s push into HBM4 also highlights intensifying competition with SK Hynix, which has been the primary supplier of HBM3 and HBM3E memory for NVIDIA’s AI accelerators. SK Hynix has benefited significantly from the AI boom over the past two years, while Samsung has faced delays in qualifying its most advanced HBM products.
Topics For More Insights
Following the Reuters report, Samsung shares rose, reflecting investor optimism that the company is gaining traction in the high-margin AI memory segment. In contrast, SK Hynix shares dipped, underscoring market sensitivity to any potential shift in NVIDIA’s supplier mix.
The broader memory industry is undergoing a structural shift as manufacturers allocate more capacity to HBM production, often at the expense of traditional DRAM. This rebalancing has been driven by sustained AI investment from cloud providers and chipmakers, reshaping earnings prospects for memory companies.
Samsung is expected to provide more details on its HBM strategy during its upcoming earnings announcements. For now, the start of HBM4 production signals that the company is positioning itself more aggressively in the race to supply the world’s most advanced AI chips.
Join The Discussion